In [1]:
cd ../..

/Users/shanekercheval/repos/data-science-template
In [2]:
%run "source/config/notebook_settings.py"
import os
import mlflow
from mlflow.tracking import MlflowClient
from helpsk.utility import read_pickle
import helpsk as hlp

from source.library.utilities import Timer, log_info, get_config

config = get_config()
mlflow_uri = config['MLFLOW']['URI']
log_info(f"MLFlow URI: {mlflow_uri}")

client = MlflowClient(tracking_uri='http://127.0.0.1:1234')

2022-06-10 18:04:11 - INFO     | MLFlow URI: http://127.0.0.1:1234

Get Latest Experiment Run from MLFlow¶

In [3]:
# Get the production model version and actual model
production_model_info = client.get_latest_versions(name=config['MLFLOW']['MODEL_NAME'], stages=['Production'])
assert len(production_model_info) == 1
production_model_info = production_model_info[0]
production_model = read_pickle(client.download_artifacts(
    run_id=production_model_info.run_id,
    path='model/model.pkl'
))
log_info(f"Production Model Version: {production_model_info.version}")

2022-06-10 18:04:11 - INFO     | Production Model Version: 1
In [4]:
# get experiment and latest run info
credit_experiment = client.get_experiment_by_name(name=config['MLFLOW']['EXPERIMENT_NAME'])
runs = client.list_run_infos(experiment_id=credit_experiment.experiment_id)
latest_run = runs[np.argmax([x.start_time for x in runs])]
In [5]:
yaml_path = client.download_artifacts(run_id=latest_run.run_id, path='experiment.yaml')
results = hlp.sklearn_eval.MLExperimentResults.from_yaml_file(yaml_file_name = yaml_path)
In [6]:
# get the best estimator from the BayesSearchCV
best_estimator = read_pickle(client.download_artifacts(
    run_id=latest_run.run_id,
    path='model/model.pkl'
))
In [7]:
best_estimator.model
Out[7]:
Pipeline(steps=[('prep',
                 ColumnTransformer(transformers=[('numeric',
                                                  Pipeline(steps=[('imputer',
                                                                   TransformerChooser(transformer=SimpleImputer(strategy='median'))),
                                                                  ('scaler',
                                                                   TransformerChooser()),
                                                                  ('pca',
                                                                   TransformerChooser(transformer=PCA(n_components='mle')))]),
                                                  ['duration', 'credit_amount',
                                                   'installment_commitment',
                                                   'residence_since', 'age',
                                                   'existing_credi...
                                                   'savings_status',
                                                   'employment',
                                                   'personal_status',
                                                   'other_parties',
                                                   'property_magnitude',
                                                   'other_payment_plans',
                                                   'housing', 'job',
                                                   'own_telephone',
                                                   'foreign_worker'])])),
                ('model',
                 RandomForestClassifier(criterion='entropy', max_depth=70,
                                        max_features=0.1142268477118407,
                                        max_samples=0.5483119512487002,
                                        min_samples_leaf=8,
                                        min_samples_split=12, n_estimators=553,
                                        random_state=42))])
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
Pipeline(steps=[('prep',
                 ColumnTransformer(transformers=[('numeric',
                                                  Pipeline(steps=[('imputer',
                                                                   TransformerChooser(transformer=SimpleImputer(strategy='median'))),
                                                                  ('scaler',
                                                                   TransformerChooser()),
                                                                  ('pca',
                                                                   TransformerChooser(transformer=PCA(n_components='mle')))]),
                                                  ['duration', 'credit_amount',
                                                   'installment_commitment',
                                                   'residence_since', 'age',
                                                   'existing_credi...
                                                   'savings_status',
                                                   'employment',
                                                   'personal_status',
                                                   'other_parties',
                                                   'property_magnitude',
                                                   'other_payment_plans',
                                                   'housing', 'job',
                                                   'own_telephone',
                                                   'foreign_worker'])])),
                ('model',
                 RandomForestClassifier(criterion='entropy', max_depth=70,
                                        max_features=0.1142268477118407,
                                        max_samples=0.5483119512487002,
                                        min_samples_leaf=8,
                                        min_samples_split=12, n_estimators=553,
                                        random_state=42))])
ColumnTransformer(transformers=[('numeric',
                                 Pipeline(steps=[('imputer',
                                                  TransformerChooser(transformer=SimpleImputer(strategy='median'))),
                                                 ('scaler',
                                                  TransformerChooser()),
                                                 ('pca',
                                                  TransformerChooser(transformer=PCA(n_components='mle')))]),
                                 ['duration', 'credit_amount',
                                  'installment_commitment', 'residence_since',
                                  'age', 'existing_credits',
                                  'num_dependents']),
                                ('non_numeric',
                                 Pipeline(steps=[('encoder',
                                                  TransformerChooser(transformer=OneHotEncoder(handle_unknown='ignore')))]),
                                 ['checking_status', 'credit_history',
                                  'purpose', 'savings_status', 'employment',
                                  'personal_status', 'other_parties',
                                  'property_magnitude', 'other_payment_plans',
                                  'housing', 'job', 'own_telephone',
                                  'foreign_worker'])])
['duration', 'credit_amount', 'installment_commitment', 'residence_since', 'age', 'existing_credits', 'num_dependents']
TransformerChooser(transformer=SimpleImputer(strategy='median'))
SimpleImputer(strategy='median')
SimpleImputer(strategy='median')
TransformerChooser()
TransformerChooser(transformer=PCA(n_components='mle'))
PCA(n_components='mle')
PCA(n_components='mle')
['checking_status', 'credit_history', 'purpose', 'savings_status', 'employment', 'personal_status', 'other_parties', 'property_magnitude', 'other_payment_plans', 'housing', 'job', 'own_telephone', 'foreign_worker']
TransformerChooser(transformer=OneHotEncoder(handle_unknown='ignore'))
OneHotEncoder(handle_unknown='ignore')
OneHotEncoder(handle_unknown='ignore')
RandomForestClassifier(criterion='entropy', max_depth=70,
                       max_features=0.1142268477118407,
                       max_samples=0.5483119512487002, min_samples_leaf=8,
                       min_samples_split=12, n_estimators=553, random_state=42)

Training & Test Data Info¶

In [8]:
client.download_artifacts(run_id=latest_run.run_id, path='x_train.pkl')
Out[8]:
'/Users/shanekercheval/repos/data-science-template/mlflow-artifact-root/1/f9ca93ecd6f14120accfec03400f6532/artifacts/x_train.pkl'
In [9]:
with Timer("Loading training/test datasets"):
    X_train = pd.pandas.read_pickle(client.download_artifacts(run_id=latest_run.run_id, path='x_train.pkl'))
    X_test = pd.pandas.read_pickle(client.download_artifacts(run_id=latest_run.run_id, path='x_test.pkl'))
    y_train = pd.pandas.read_pickle(client.download_artifacts(run_id=latest_run.run_id, path='y_train.pkl'))
    y_test = pd.pandas.read_pickle(client.download_artifacts(run_id=latest_run.run_id, path='y_test.pkl'))

2022-06-10 18:04:11 - INFO     | *****Timer Started: Loading training/test datasets
2022-06-10 18:04:11 - INFO     | *****Timer Finished (0.00 seconds)
In [10]:
log_info(X_train.shape)
log_info(len(y_train))

log_info(X_test.shape)
log_info(len(y_test))

2022-06-10 18:04:11 - INFO     | (800, 20)
2022-06-10 18:04:11 - INFO     | 800
2022-06-10 18:04:11 - INFO     | (200, 20)
2022-06-10 18:04:11 - INFO     | 200
In [11]:
np.unique(y_train, return_counts=True)
Out[11]:
(array([0, 1]), array([559, 241]))
In [12]:
np.unique(y_train, return_counts=True)[1] / np.sum(np.unique(y_train, return_counts=True)[1])
Out[12]:
array([0.69875, 0.30125])
In [13]:
np.unique(y_test, return_counts=True)[1] / np.sum(np.unique(y_test, return_counts=True)[1])
Out[13]:
array([0.705, 0.295])

Cross Validation Results¶

Best Scores/Params¶

In [14]:
log_info(f"Best Score: {results.best_score}")

2022-06-10 18:04:11 - INFO     | Best Score: 0.7626843675980791
In [15]:
log_info(f"Best Params: {results.best_params}")

2022-06-10 18:04:11 - INFO     | Best Params: {'model': 'RandomForestClassifier()', 'max_features': 0.1142268477118407, 'max_depth': 70, 'n_estimators': 553, 'min_samples_split': 12, 'min_samples_leaf': 8, 'max_samples': 0.5483119512487002, 'criterion': 'entropy', 'imputer': "SimpleImputer(strategy='median')", 'scaler': 'None', 'pca': "PCA('mle')", 'encoder': 'OneHotEncoder()'}
In [16]:
# Best model from each model-type.
df = results.to_formatted_dataframe(return_style=False, include_rank=True)
df["model_rank"] = df.groupby("model")["roc_auc Mean"].rank(method="first", ascending=False)
df.query('model_rank == 1')
Out[16]:
rank roc_auc Mean roc_auc 95CI.LO roc_auc 95CI.HI model C max_features max_depth n_estimators min_samples_split min_samples_leaf max_samples criterion learning_rate min_child_weight subsample colsample_bytree colsample_bylevel reg_alpha reg_lambda imputer scaler pca encoder model_rank
13 1 0.76 0.71 0.81 RandomForestClassifier() NaN 0.11 70.00 553.00 12.00 8.00 0.55 entropy NaN NaN NaN NaN NaN NaN NaN SimpleImputer(strategy='median') None PCA('mle') OneHotEncoder() 1.00
0 3 0.76 0.71 0.81 LogisticRegression() NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN SimpleImputer() StandardScaler() None OneHotEncoder() 1.00
7 4 0.76 0.71 0.80 ExtraTreesClassifier() NaN 0.70 38.00 558.00 18.00 1.00 0.51 gini NaN NaN NaN NaN NaN NaN NaN SimpleImputer() None PCA('mle') OneHotEncoder() 1.00
19 11 0.73 0.70 0.76 XGBClassifier() NaN NaN 15.00 1159.00 NaN NaN NaN NaN 0.03 29.00 0.83 0.52 0.50 0.00 1.84 SimpleImputer(strategy='median') None PCA('mle') CustomOrdinalEncoder() 1.00
In [17]:
results.to_formatted_dataframe(return_style=True,
                               include_rank=True,
                               num_rows=500)
Out[17]:
rank roc_auc Mean roc_auc 95CI.LO roc_auc 95CI.HI model C max_features max_depth n_estimators min_samples_split min_samples_leaf max_samples criterion learning_rate min_child_weight subsample colsample_bytree colsample_bylevel reg_alpha reg_lambda imputer scaler pca encoder
1 0.763 0.712 0.814 RandomForestClassifier() <NA> 0.114 70.000 553.000 12.000 8.000 0.548 entropy <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') None PCA('mle') OneHotEncoder()
2 0.759 0.699 0.819 RandomForestClassifier() <NA> 0.030 84.000 1,088.000 24.000 36.000 0.981 gini <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() None None OneHotEncoder()
3 0.759 0.713 0.805 LogisticRegression() <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() StandardScaler() None OneHotEncoder()
4 0.759 0.714 0.804 ExtraTreesClassifier() <NA> 0.700 38.000 558.000 18.000 1.000 0.514 gini <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() None PCA('mle') OneHotEncoder()
5 0.756 0.726 0.786 RandomForestClassifier() <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() None None OneHotEncoder()
6 0.752 0.684 0.819 LogisticRegression() 0.001 <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') MinMaxScaler() None OneHotEncoder()
7 0.744 0.709 0.780 RandomForestClassifier() <NA> 0.681 38.000 1,461.000 23.000 10.000 0.553 gini <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') None None CustomOrdinalEncoder()
8 0.742 0.689 0.796 ExtraTreesClassifier() <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() None None OneHotEncoder()
9 0.742 0.703 0.781 ExtraTreesClassifier() <NA> 0.473 25.000 834.000 48.000 6.000 0.631 gini <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() None PCA('mle') CustomOrdinalEncoder()
10 0.741 0.689 0.793 RandomForestClassifier() <NA> 0.559 72.000 1,330.000 35.000 32.000 0.630 gini <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='most_frequent') None PCA('mle') OneHotEncoder()
11 0.726 0.698 0.755 XGBClassifier() <NA> <NA> 15.000 1,159.000 <NA> <NA> <NA> <NA> 0.032 29.000 0.834 0.520 0.503 0.003 1.839 SimpleImputer(strategy='median') None PCA('mle') CustomOrdinalEncoder()
12 0.726 0.689 0.762 LogisticRegression() 0.000 <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') StandardScaler() None CustomOrdinalEncoder()
13 0.723 0.637 0.809 LogisticRegression() 0.403 <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') StandardScaler() None CustomOrdinalEncoder()
14 0.723 0.666 0.779 ExtraTreesClassifier() <NA> 0.857 30.000 879.000 17.000 28.000 0.563 entropy <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') None None CustomOrdinalEncoder()
15 0.722 0.658 0.786 ExtraTreesClassifier() <NA> 0.781 50.000 590.000 35.000 47.000 0.846 gini <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') None PCA('mle') OneHotEncoder()
16 0.720 0.699 0.741 XGBClassifier() <NA> <NA> 19.000 1,217.000 <NA> <NA> <NA> <NA> 0.042 4.000 0.893 0.936 0.652 0.029 3.708 SimpleImputer(strategy='most_frequent') None None CustomOrdinalEncoder()
17 0.720 0.638 0.801 LogisticRegression() 3.489 <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer(strategy='median') StandardScaler() PCA('mle') CustomOrdinalEncoder()
18 0.718 0.695 0.740 XGBClassifier() <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() None None OneHotEncoder()
19 0.716 0.666 0.766 XGBClassifier() <NA> <NA> 12.000 1,471.000 <NA> <NA> <NA> <NA> 0.126 8.000 0.912 0.569 0.821 0.045 1.029 SimpleImputer() None PCA('mle') CustomOrdinalEncoder()
20 0.711 0.659 0.764 XGBClassifier() <NA> <NA> 5.000 1,218.000 <NA> <NA> <NA> <NA> 0.115 2.000 0.545 0.648 0.852 0.123 1.165 SimpleImputer(strategy='median') None PCA('mle') CustomOrdinalEncoder()
In [18]:
results.to_formatted_dataframe(query='model == "RandomForestClassifier()"', include_rank=True)
Out[18]:
rank roc_auc Mean roc_auc 95CI.LO roc_auc 95CI.HI max_features max_depth n_estimators min_samples_split min_samples_leaf max_samples criterion imputer pca encoder
1 0.763 0.712 0.814 0.114 70.000 553.000 12.000 8.000 0.548 entropy SimpleImputer(strategy='median') PCA('mle') OneHotEncoder()
2 0.759 0.699 0.819 0.030 84.000 1,088.000 24.000 36.000 0.981 gini SimpleImputer() None OneHotEncoder()
3 0.756 0.726 0.786 <NA> <NA> <NA> <NA> <NA> <NA> <NA> SimpleImputer() None OneHotEncoder()
4 0.744 0.709 0.780 0.681 38.000 1,461.000 23.000 10.000 0.553 gini SimpleImputer(strategy='median') None CustomOrdinalEncoder()
5 0.741 0.689 0.793 0.559 72.000 1,330.000 35.000 32.000 0.630 gini SimpleImputer(strategy='most_frequent') PCA('mle') OneHotEncoder()
In [19]:
results.to_formatted_dataframe(query='model == "LogisticRegression()"', include_rank=True)
Out[19]:
rank roc_auc Mean roc_auc 95CI.LO roc_auc 95CI.HI C imputer scaler pca encoder
1 0.759 0.713 0.805 <NA> SimpleImputer() StandardScaler() None OneHotEncoder()
2 0.752 0.684 0.819 0.001 SimpleImputer(strategy='median') MinMaxScaler() None OneHotEncoder()
3 0.726 0.689 0.762 0.000 SimpleImputer(strategy='median') StandardScaler() None CustomOrdinalEncoder()
4 0.723 0.637 0.809 0.403 SimpleImputer(strategy='median') StandardScaler() None CustomOrdinalEncoder()
5 0.720 0.638 0.801 3.489 SimpleImputer(strategy='median') StandardScaler() PCA('mle') CustomOrdinalEncoder()

BayesSearchCV Performance Over Time¶

In [20]:
results.plot_performance_across_trials(facet_by='model').show()
In [21]:
results.plot_performance_across_trials(query='model == "RandomForestClassifier()"').show()

Variable Performance Over Time¶

In [22]:
results.plot_parameter_values_across_trials(query='model == "RandomForestClassifier()"').show()

Scatter Matrix¶

In [23]:
# results.plot_scatter_matrix(query='model == "RandomForestClassifier()"',
#                             height=1000, width=1000).show()

Variable Performance - Numeric¶

In [24]:
results.plot_performance_numeric_params(query='model == "RandomForestClassifier()"',
                                        height=800)
In [25]:
results.plot_parallel_coordinates(query='model == "RandomForestClassifier()"').show()

Variable Performance - Non-Numeric¶

In [26]:
results.plot_performance_non_numeric_params(query='model == "RandomForestClassifier()"').show()
In [27]:
results.plot_score_vs_parameter(
    query='model == "RandomForestClassifier()"',
    parameter='max_features',
    size='max_depth',
    color='encoder',
)
In [28]:
# results.plot_parameter_vs_parameter(
#     query='model == "XGBClassifier()"',
#     parameter_x='colsample_bytree',
#     parameter_y='learning_rate',
#     size='max_depth'
# )
In [29]:
# results.plot_parameter_vs_parameter(
#     query='model == "XGBClassifier()"',
#     parameter_x='colsample_bytree',
#     parameter_y='learning_rate',
#     size='imputer'
# )

Best Model - Test Set Performance¶

In [30]:
test_predictions = best_estimator.predict(X_test)
test_predictions[0:10]
Out[30]:
array([0.37346429, 0.4104999 , 0.48390757, 0.35895319, 0.17376038,
       0.32902654, 0.17785134, 0.40538264, 0.21201897, 0.24403248])
In [31]:
evaluator = hlp.sklearn_eval.TwoClassEvaluator(
    actual_values=y_test,
    predicted_scores=test_predictions,
    score_threshold=0.37
)
In [32]:
evaluator.plot_actual_vs_predict_histogram()
In [33]:
evaluator.plot_confusion_matrix()
In [34]:
evaluator.all_metrics_df(return_style=True,
                         dummy_classifier_strategy=['prior', 'constant'],
                         round_by=3)
Out[34]:
  Score Dummy (prior) Dummy (constant) Explanation
AUC 0.794 0.500 0.500 Area under the ROC curve (true pos. rate vs false pos. rate); ranges from 0.5 (purely random classifier) to 1.0 (perfect classifier)
True Positive Rate 0.593 0.000 1.000 59.3% of positive instances were correctly identified.; i.e. 35 "Positive Class" labels were correctly identified out of 59 instances; a.k.a Sensitivity/Recall
True Negative Rate 0.837 1.000 0.000 83.7% of negative instances were correctly identified.; i.e. 118 "Negative Class" labels were correctly identified out of 141 instances
False Positive Rate 0.163 0.000 1.000 16.3% of negative instances were incorrectly identified as positive; i.e. 23 "Negative Class" labels were incorrectly identified as "Positive Class", out of 141 instances
False Negative Rate 0.407 1.000 0.000 40.7% of positive instances were incorrectly identified as negative; i.e. 24 "Positive Class" labels were incorrectly identified as "Negative Class", out of 59 instances
Positive Predictive Value 0.603 0.000 0.295 When the model claims an instance is positive, it is correct 60.3% of the time; i.e. out of the 58 times the model predicted "Positive Class", it was correct 35 times; a.k.a precision
Negative Predictive Value 0.831 0.705 0.000 When the model claims an instance is negative, it is correct 83.1% of the time; i.e. out of the 142 times the model predicted "Negative Class", it was correct 118 times
F1 Score 0.598 0.000 0.456 The F1 score can be interpreted as a weighted average of the precision and recall, where an F1 score reaches its best value at 1 and worst score at 0.
Precision/Recall AUC 0.642 0.295 0.295 Precision/Recall AUC is calculated with `average_precision` which summarizes a precision-recall curve as the weighted mean of precisions achieved at each threshold. See sci-kit learn documentation for caveats.
Accuracy 0.765 0.705 0.295 76.5% of instances were correctly identified
Error Rate 0.235 0.295 0.705 23.5% of instances were incorrectly identified
% Positive 0.295 0.295 0.295 29.5% of the data are positive; i.e. out of 200 total observations; 59 are labeled as "Positive Class"
Total Observations 200 200 200 There are 200 total observations; i.e. sample size
In [35]:
evaluator.plot_roc_auc_curve().show()

<Figure size 720x444.984 with 0 Axes>
In [36]:
evaluator.plot_precision_recall_auc_curve().show()
In [37]:
evaluator.plot_threshold_curves(score_threshold_range=(0.1, 0.7)).show()
In [38]:
evaluator.plot_precision_recall_tradeoff(score_threshold_range=(0.1, 0.6)).show()
In [39]:
evaluator.calculate_lift_gain(return_style=True)
Out[39]:
  Gain Lift
Percentile    
5 0.17 3.39
10 0.25 2.54
15 0.32 2.15
20 0.42 2.12
25 0.53 2.10
30 0.61 2.03
35 0.68 1.94
40 0.75 1.86
45 0.76 1.69
50 0.80 1.59
55 0.80 1.45
60 0.88 1.47
65 0.88 1.36
70 0.92 1.31
75 0.93 1.24
80 0.97 1.21
85 0.98 1.16
90 1.00 1.11
95 1.00 1.05
100 1.00 1.00

Production Model - Test Set Performance¶

In [40]:
test_predictions = production_model.predict(X_test)
test_predictions[0:10]
Out[40]:
array([0.402, 0.494, 0.722, 0.374, 0.056, 0.476, 0.084, 0.476, 0.18 ,
       0.232])
In [41]:
evaluator = hlp.sklearn_eval.TwoClassEvaluator(
    actual_values=y_test,
    predicted_scores=test_predictions,
    score_threshold=0.37
)
In [42]:
evaluator.plot_actual_vs_predict_histogram()
In [43]:
evaluator.plot_confusion_matrix()
In [44]:
evaluator.all_metrics_df(return_style=True,
                         dummy_classifier_strategy=['prior', 'constant'],
                         round_by=3)
Out[44]:
  Score Dummy (prior) Dummy (constant) Explanation
AUC 0.825 0.500 0.500 Area under the ROC curve (true pos. rate vs false pos. rate); ranges from 0.5 (purely random classifier) to 1.0 (perfect classifier)
True Positive Rate 0.746 0.000 1.000 74.6% of positive instances were correctly identified.; i.e. 44 "Positive Class" labels were correctly identified out of 59 instances; a.k.a Sensitivity/Recall
True Negative Rate 0.801 1.000 0.000 80.1% of negative instances were correctly identified.; i.e. 113 "Negative Class" labels were correctly identified out of 141 instances
False Positive Rate 0.199 0.000 1.000 19.9% of negative instances were incorrectly identified as positive; i.e. 28 "Negative Class" labels were incorrectly identified as "Positive Class", out of 141 instances
False Negative Rate 0.254 1.000 0.000 25.4% of positive instances were incorrectly identified as negative; i.e. 15 "Positive Class" labels were incorrectly identified as "Negative Class", out of 59 instances
Positive Predictive Value 0.611 0.000 0.295 When the model claims an instance is positive, it is correct 61.1% of the time; i.e. out of the 72 times the model predicted "Positive Class", it was correct 44 times; a.k.a precision
Negative Predictive Value 0.883 0.705 0.000 When the model claims an instance is negative, it is correct 88.3% of the time; i.e. out of the 128 times the model predicted "Negative Class", it was correct 113 times
F1 Score 0.672 0.000 0.456 The F1 score can be interpreted as a weighted average of the precision and recall, where an F1 score reaches its best value at 1 and worst score at 0.
Precision/Recall AUC 0.667 0.295 0.295 Precision/Recall AUC is calculated with `average_precision` which summarizes a precision-recall curve as the weighted mean of precisions achieved at each threshold. See sci-kit learn documentation for caveats.
Accuracy 0.785 0.705 0.295 78.5% of instances were correctly identified
Error Rate 0.215 0.295 0.705 21.5% of instances were incorrectly identified
% Positive 0.295 0.295 0.295 29.5% of the data are positive; i.e. out of 200 total observations; 59 are labeled as "Positive Class"
Total Observations 200 200 200 There are 200 total observations; i.e. sample size
In [45]:
evaluator.plot_roc_auc_curve().show()

<Figure size 720x444.984 with 0 Axes>
In [46]:
evaluator.plot_precision_recall_auc_curve().show()
In [47]:
evaluator.plot_threshold_curves(score_threshold_range=(0.1, 0.7)).show()
In [48]:
evaluator.plot_precision_recall_tradeoff(score_threshold_range=(0.1, 0.6)).show()
In [49]:
evaluator.calculate_lift_gain(return_style=True)
Out[49]:
  Gain Lift
Percentile    
5 0.14 2.71
10 0.24 2.37
15 0.37 2.49
20 0.49 2.46
25 0.56 2.24
30 0.64 2.15
35 0.75 2.13
40 0.76 1.91
45 0.80 1.77
50 0.83 1.66
55 0.85 1.54
60 0.88 1.47
65 0.88 1.36
70 0.93 1.33
75 0.95 1.27
80 0.97 1.21
85 0.98 1.16
90 1.00 1.11
95 1.00 1.05
100 1.00 1.00